Phantom // Dev Blog March 2026

Developer Guide Web3 Wallets AI Agents

Build anything with Phantom's developer platform

From embedded wallets in mobile apps to AI agents trading on-chain — here's everything developers need to know to get started with Phantom.

Developer Docs Deep-Dive · 8 min read · March 29, 2026

The Big Picture

Phantom isn't just a wallet anymore

Most developers know Phantom as the dominant browser extension and mobile wallet for Solana. But the product has grown far beyond that. Today, Phantom supports eight blockchains and offers a full developer platform — giving you the tools to embed wallet functionality directly into your apps, or even hand a wallet to an AI agent.

Whether you're building a DeFi dashboard, a gaming app, or an autonomous trading bot, there's a Phantom integration path designed for your use case.

Solana

Ethereum

Bitcoin

Base

Polygon

Sui

Monad

HyperEVM

Multi-chain support isn't an afterthought — it's baked into all of Phantom's SDKs from day one. Your users can switch chains without leaving your app.

Two Paths

Who are you building for?

Phantom's developer platform is organized around a simple question: who's the end user? A human using your app, or an AI agent acting autonomously? The answer determines which integration you reach for.

"Give AI agents a Phantom wallet. Let humans have one too."

For human users, you integrate via the Phantom Connect SDKs — React, React Native, or a framework-agnostic Browser SDK. These give you embedded wallets, social login, and native transaction signing right inside your UI.

For AI agents, you use the Phantom MCP Server — a brand-new product that lets any agent with MCP support hold and operate a real Phantom wallet on-chain.

For Human Users

The Phantom Connect SDK suite

Phantom offers three SDKs that cover the full spectrum of web and mobile development. All of them share the same underlying multi-chain architecture — you pick the one that fits your stack.

React SDK

Hooks-based integration for React web apps. Native transaction support out of the box.

React Native SDK

Build mobile wallet experiences for iOS and Android. Same API, native feel.

Browser SDK

Framework-agnostic JavaScript. Works with Vue, Svelte, vanilla JS — anything.

Not sure which to use? Phantom provides an SDK comparison guide. As a rule of thumb: reach for the React SDK if you're building a standard web app, React Native if you're targeting mobile, and the Browser SDK if you need flexibility across frameworks.

// React SDK — connect a wallet in two lines import { useWallet } from '@phantom/react'; const { connect, publicKey } = useWallet(); await connect(); // Opens Phantom, returns public key

For AI Agents

The Phantom MCP Server

This is arguably the most exciting thing Phantom has shipped for developers recently. The MCP Server is a first-class Phantom product — think of it as the browser extension, but for AI agents.

What agents can do

📤 Send transactions and sign messages on Solana and EVM chains

💱 Swap tokens — including cross-chain swaps

📊 View balances and rebalance portfolios

🔑 Manage wallet addresses across all supported networks

The MCP Server exposes 13 tools across two capability groups: wallet operations and swaps/portfolio management. Any MCP-compatible agent runtime can connect — including Claude Desktop, Cursor, and Claude Code.

The setup process is straightforward: add the Phantom MCP Server to your agent environment and it handles key management, transaction construction, and chain interactions automatically. Your agent's prompt stays focused on intent, not implementation details.

This opens up genuinely new product categories — autonomous DeFi strategies, AI-assisted portfolio management, or agents that bridge assets across chains on behalf of users.

Getting Started

Resources to hit the ground running

Phantom's docs include a solid set of resources for developers beyond the core SDK and MCP documentation:

Recipes and starter kits give you copy-paste templates for common integration patterns — useful for getting a working prototype running in minutes rather than hours.

The interactive sandbox lets you test your integration live without deploying anything, which is a significant time-saver when you're iterating on transaction flows.

For production support, Phantom has a dedicated developer support request form — notably separate from their general consumer help center, which signals they take the developer experience seriously.

The sandbox alone will save you hours of debug cycles. Test your full transaction flow before writing a single line of app code.

All documentation is available at docs.phantom.com, and there's an llms.txt index at the root for programmatic navigation — handy if you're building tooling on top of the docs.

Bottom Line

What this means for builders

Phantom's developer platform has reached a level of maturity where it's genuinely competitive with any wallet SDK on the market. The multi-chain support, the three-SDK strategy, and especially the MCP Server make it a compelling choice whether you're building a consumer app or an autonomous agent.

The MCP Server in particular is a forward-looking bet that AI agents will increasingly need to interact with on-chain assets — and Phantom is positioning itself as the infrastructure layer for that future. If you're building in the AI x crypto space, it's worth experimenting with now.

Start with the docs, run through the sandbox, and pick the SDK that matches your stack. The path from zero to a working wallet integration is shorter than it's ever been.

Built with Phantom developer docs · docs.phantom.com

 

Can Convolutional Neural Networks Predict Stock Prices?

🧠 What is a CNN (in simple terms)?

A Convolutional Neural Network (CNN) is a type of deep learning model designed to automatically detect patterns in data.

• Originally built for images (like recognizing faces)
• Works by scanning data using filters to detect patterns (edges, shapes → complex features)

👉 Think of it like this:

• Early layers = detect simple patterns
• Deeper layers = combine them into meaningful signals

---

📈 How CNNs are used in the stock market

Even though CNNs are famous for images, they can also analyze financial data by converting it into structured formats.

1. Input data (what goes into the CNN)

CNNs don’t just take raw prices—they often use:

• Historical stock prices
• Technical indicators (RSI, moving averages)
• Trading volume
• Even chart images or transformed data

These can be turned into:

• Time-series matrices
• Heatmaps / images of charts
• Multi-channel datasets (like RGB images but for finance)

👉 This allows CNNs to “see” patterns in market behavior

---

2. What the CNN learns

CNNs automatically detect:

• Short-term price patterns
• Trend formations
• Repeating structures in data

This is powerful because:

• Markets are nonlinear and noisy
• Traditional models struggle with this complexity

---

3. Typical tasks in trading

CNNs are used for:

• 📊 Price prediction (next price / return)
• 🔼 Trend classification (up, down, sideways)
• 💡 Trading signals (buy / sell / hold)
• 📉 Volatility or risk prediction

---

⚙️ Why CNNs work well for stocks

Key advantages:

• Pattern detection → finds local trends in price movements
• Feature learning → no need to manually design indicators
• Scalability → can analyze many stocks at once

👉 Some studies report improved accuracy over traditional models (like ARIMA or regression)

---

⚠️ Important reality check

CNNs are powerful—but not magic:

• Stock markets are highly unpredictable
• Influenced by news, psychology, macroeconomics
• Models can overfit (look good in past data, fail in real trading)

👉 So CNNs are used as decision-support tools, not guaranteed profit machines.

---

🧩 Simple analogy

Think of a CNN like a technical analyst on steroids:

• A human sees patterns in charts
• A CNN sees thousands of patterns at once—and learns which ones matter

---

✅ Bottom line

• CNNs = AI models that detect patterns
• In stocks, they analyze price data like images
• They help predict trends and signals
• Useful, but not foolproof

---

If you want, I can show you:

• A simple Python example of CNN for stock prediction
• Or compare CNN vs LSTM vs traditional models (very useful for interviews/projects)